Upload fine-tuned model - BLEU: 0.1277

.gitattributes

@@ -43,3 +43,5 @@ source.spm filter=lfs diff=lfs merge=lfs -text
 target.spm filter=lfs diff=lfs merge=lfs -text
 checkpoint-128/source.spm filter=lfs diff=lfs merge=lfs -text
 checkpoint-128/target.spm filter=lfs diff=lfs merge=lfs -text
+checkpoint-96/source.spm filter=lfs diff=lfs merge=lfs -text
+checkpoint-96/target.spm filter=lfs diff=lfs merge=lfs -text

README.md

@@ -40,13 +40,13 @@ The model was fine-tuned on a combination of:
 ## Performance
 
 ### Test Set (General Translation)
-- **BLEU:** 0.1556
-- **chrF:** 35.07
+- **BLEU:** 0.1277
+- **chrF:** 32.30
 - **Improvement over baseline:** +0.0%
 
 ### Domain Evaluation (Call Transcriptions)
 - **Domain BLEU:** 0.0000
-- **Domain chrF:** 2.19
+- **Domain chrF:** 1.86
 - **Domain COMET-QE:** 0.0000
 
 *Domain metrics are evaluated on real 10-minute call transcriptions from child helplines.*
@@ -88,7 +88,7 @@ print(translation)
 
 **Base Model:** Helsinki-NLP/opus-mt-mul-en  
 **Training Epochs:** 8  
-**Batch Size:** 400 (effective: 400)  
+**Batch Size:** 550 (effective: 550)  
 **Learning Rate:** 3e-05  
 **Optimizer:** AdamW  
 **Hardware:** NVIDIA GPU with FP16 mixed precision

baseline_results.json

@@ -1,10 +1,10 @@
 {
   "eval_loss": 18.773775100708008,
-  "eval_model_preparation_time": 0.0024,
+  "eval_model_preparation_time": 0.0027,
   "eval_bleu": 0.0317523275253792,
   "eval_chrf": 19.994361360713377,
-  "eval_runtime": 18.6261,
-  "eval_samples_per_second": 19.328,
+  "eval_runtime": 18.4416,
+  "eval_samples_per_second": 19.521,
   "eval_steps_per_second": 0.054,
   "eval_comet": 0.0341746505332392,
   "eval_comet_std": 0.1127765594779826,

checkpoint-96/config.json

@@ -0,0 +1,54 @@
+{
+  "activation_dropout": 0.0,
+  "activation_function": "swish",
+  "add_bias_logits": false,
+  "add_final_layer_norm": false,
+  "architectures": [
+    "MarianMTModel"
+  ],
+  "attention_dropout": 0.0,
+  "classif_dropout": 0.0,
+  "classifier_dropout": 0.0,
+  "d_model": 512,
+  "decoder_attention_heads": 8,
+  "decoder_ffn_dim": 2048,
+  "decoder_layerdrop": 0.0,
+  "decoder_layers": 6,
+  "decoder_start_token_id": 64171,
+  "decoder_vocab_size": 64172,
+  "dropout": 0.1,
+  "dtype": "float32",
+  "encoder_attention_heads": 8,
+  "encoder_ffn_dim": 2048,
+  "encoder_layerdrop": 0.0,
+  "encoder_layers": 6,
+  "eos_token_id": 0,
+  "extra_pos_embeddings": 64172,
+  "forced_eos_token_id": 0,
+  "id2label": {
+    "0": "LABEL_0",
+    "1": "LABEL_1",
+    "2": "LABEL_2"
+  },
+  "init_std": 0.02,
+  "is_encoder_decoder": true,
+  "label2id": {
+    "LABEL_0": 0,
+    "LABEL_1": 1,
+    "LABEL_2": 2
+  },
+  "max_length": null,
+  "max_position_embeddings": 512,
+  "model_type": "marian",
+  "normalize_before": false,
+  "normalize_embedding": false,
+  "num_beams": null,
+  "num_hidden_layers": 6,
+  "pad_token_id": 64171,
+  "scale_embedding": true,
+  "share_encoder_decoder_embeddings": true,
+  "static_position_embeddings": true,
+  "transformers_version": "4.57.1",
+  "use_cache": true,
+  "vocab_size": 64172
+}

checkpoint-96/generation_config.json

@@ -0,0 +1,17 @@
+{
+  "bad_words_ids": [
+    [
+      64171
+    ]
+  ],
+  "decoder_start_token_id": 64171,
+  "eos_token_id": [
+    0
+  ],
+  "forced_eos_token_id": 0,
+  "max_length": 512,
+  "num_beams": 6,
+  "pad_token_id": 64171,
+  "renormalize_logits": true,
+  "transformers_version": "4.57.1"
+}

checkpoint-96/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4e1c9e6b6073e8a0fc83c1a11432fbadf83a5c32033b36823091975ede4c568f
+size 308263984

checkpoint-96/optimizer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:98a43be8f8d6ec5dc6739423b9744963df61bcb70c99ace839fc38350e209a2c
+size 616171979

checkpoint-96/rng_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:abe68b8d07bb856812fa822df28c97b9df77fbbe797028ee7cd9f8e7dd846374
+size 14645

checkpoint-96/scaler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:68a92e81e21ef52284b52c64167c5820c17700eece1dc77ea3e5c9144c32e6c5
+size 1383

checkpoint-96/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3b12b4be9f022ed85c83fe8574cc4bbe781b092616f5f3b78c3b587e32292c49
+size 1465

checkpoint-96/source.spm

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c4a99ea3602b29fbf901ade8b93a45efa3d7c64eab8fc5fa812383efa327a87d
+size 706917

checkpoint-96/special_tokens_map.json

@@ -0,0 +1,5 @@
+{
+  "eos_token": "</s>",
+  "pad_token": "<pad>",
+  "unk_token": "<unk>"
+}

checkpoint-96/target.spm

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c6dce5fa58fcd7dde9e81e279b8c075bf42ee558278f73d6fb48e342029d7f19
+size 791194

checkpoint-96/tokenizer_config.json

@@ -0,0 +1,39 @@
+{
+  "added_tokens_decoder": {
+    "0": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "64171": {
+      "content": "<pad>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "</s>",
+  "extra_special_tokens": {},
+  "model_max_length": 512,
+  "pad_token": "<pad>",
+  "separate_vocabs": false,
+  "source_lang": "mul",
+  "sp_model_kwargs": {},
+  "target_lang": "eng",
+  "tokenizer_class": "MarianTokenizer",
+  "unk_token": "<unk>"
+}

checkpoint-96/trainer_state.json

@@ -0,0 +1,50 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 8.0,
+  "eval_steps": 1000,
+  "global_step": 96,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 4.166666666666667,
+      "grad_norm": 2.2876839637756348,
+      "learning_rate": 1.7184017516025075e-05,
+      "loss": 6.4599,
+      "step": 50
+    }
+  ],
+  "logging_steps": 50,
+  "max_steps": 96,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 8,
+  "save_steps": 1000,
+  "stateful_callbacks": {
+    "EarlyStoppingCallback": {
+      "args": {
+        "early_stopping_patience": 5,
+        "early_stopping_threshold": 0.001
+      },
+      "attributes": {
+        "early_stopping_patience_counter": 0
+      }
+    },
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1138408789377024.0,
+  "train_batch_size": 550,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-96/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eb0b9e675407ca1a9ee1a08d6b6b650e16ca2de15fd3b2c37137b42aedec39fa
+size 5969

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:87fda119e434228efc5a1b0d407187d15822ee257fe907afc5bf0e776aff51a5
+oid sha256:4e1c9e6b6073e8a0fc83c1a11432fbadf83a5c32033b36823091975ede4c568f
 size 308263984

test_results.json

@@ -1,16 +1,16 @@
 {
-  "eval_loss": 3.379178524017334,
-  "eval_model_preparation_time": 0.0024,
-  "eval_bleu": 0.15563550589842806,
-  "eval_chrf": 35.07391489522072,
-  "eval_runtime": 3.9553,
-  "eval_samples_per_second": 91.017,
-  "eval_steps_per_second": 0.253,
+  "eval_loss": 3.5031542778015137,
+  "eval_model_preparation_time": 0.0027,
+  "eval_bleu": 0.1276835459626037,
+  "eval_chrf": 32.29660565619225,
+  "eval_runtime": 4.0246,
+  "eval_samples_per_second": 89.449,
+  "eval_steps_per_second": 0.248,
   "epoch": 8.0,
-  "eval_comet": 0.10842741441648387,
-  "eval_comet_std": 0.2045974078652866,
-  "domain_bleu": 1.497364274347382e-14,
-  "domain_chrf": 2.1940501418043463,
-  "domain_comet": 4.183274068054743e-05,
-  "domain_comet_std": 3.4854151598511357e-07
+  "eval_comet": 0.08929400351530098,
+  "eval_comet_std": 0.1904605300394193,
+  "domain_bleu": 8.850691462561237e-18,
+  "domain_chrf": 1.8573826171296468,
+  "domain_comet": 4.175902431597933e-05,
+  "domain_comet_std": 3.194240118007998e-07
 }

train_results.json

@@ -1,8 +1,8 @@
 {
-  "train_runtime": 60.4578,
-  "train_samples_per_second": 807.969,
-  "train_steps_per_second": 2.117,
+  "train_runtime": 62.295,
+  "train_samples_per_second": 784.14,
+  "train_steps_per_second": 1.541,
   "total_flos": 1138408789377024.0,
-  "train_loss": 4.656409025192261,
+  "train_loss": 5.05125633875529,
   "epoch": 8.0
 }

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:afbf6ae69192cec42d54df2b3664029811aa8495516da29f92e1490e0a3555aa
+oid sha256:eb0b9e675407ca1a9ee1a08d6b6b650e16ca2de15fd3b2c37137b42aedec39fa
 size 5969

translation_config.json

@@ -30,7 +30,7 @@
   },
   "training_config": {
     "learning_rate": 3e-05,
-    "batch_size": 400,
+    "batch_size": 550,
     "gradient_accumulation_steps": 1,
     "num_epochs": 8,
     "max_length": 300,